Fedora image failed to setup swarm cluster

[Kai Qiang Wu(Kennan)]

I tried the fedora swarm bay, and fould it can not work at all,

I login to the instance and run docker logs check, and found that

time="2015-06-19T10:29:10Z" level=error msg="engine 10.0.0.4:2375 is running an unsupported version of Docker Engine. Please upgrade to at least 1.6.0"
time="2015-06-19T10:29:10Z" level=error msg="engine 10.0.0.3:2375 is running an unsupported version of Docker Engine. Please upgrade to at least 1.6.0"

Client version: 1.5.0
Client API version: 1.17
Go version (client): go1.3.3
Git commit (client): a8a31ef/1.5.0
OS/Arch (client): linux/amd64
Server version: 1.5.0
Server API version: 1.17
Go version (server): go1.3.3
Git commit (server): a8a31ef/1.5.0

It is that our docker engineer(the image built with) is 1.5.0 which is old, not supported in swarm,
which cause the swarm cluster not work at all.

[Kai Qiang Wu(Kennan) (wkqwu)]

Add @Sdake, as he is familar with how to build new image.

Add @Andrew, as he did some work for docker in heat enablement.

If you two not have time to build new image, I'd like to try. But right now, I did not know the steps to do that.

Thanks

[Kai Qiang Wu(Kennan) (wkqwu)]

Also seems fedora 21 have issues docker 1.6.0

https://github.com/docker/swarm/issues/781

So right now, we may not use that. I am continue investigating that

[Andrew Melton (andrew-melton)]

Hey Kennan, I've been testing this out and I've hit both of your issues and a couple more.

What I've tested:
1) fedora-21-atomic, using swarm:latest image for swarm-agent and swarm-manager containers
2) fedora-21-atomic, using swarm:0.2.0 image for swarm-agent and swarm-manager containers
3) fedora-22-atomic, using swarm:latest image for swarm-containers
4) fedora-22-atomic, using dockerswarm/swarm image for swarm-containers

Here are the issues I've hit in each case:
1) unsupported version of Docker engine
2) no issues, container-create, container-list, container-start, container-logs seem to be working
3) hit original issue from https://github.com/docker/swarm/issues/781
4) swarm containers started, container create wouldn't work with 'cirros' image, had to use 'docker.io/cirros', this works, but then container list appears to break.

I've also done some digging on the last two issues:
3) This appears to be caused by how 1.6.0 handles images built with 'FROM scratch', I can build a swarm image with FROM debian and it works fine, same with the dockerswarm/swarm image build with FROM golang. Could have something to do with the xfs driver.
4) It appears there might be some odd interactions between swarm and the image string including 'docker.io.' I can quite figure out where that's coming from. I don't see it when I install 1.6.0 on a test box...

For now, I think our best option is to pin the swarm containers at 0.2.0 so that the existing template works with our current image. Then start working towards fixing the issues we hit with fedora-22-atomic. But, I'd definitely like some input on that decision.

[Kai Qiang Wu(Kennan) (wkqwu)]

Yes, I think pin to 0.2.0 is better right now, As it is easy to rollback, not effect any other function in magnum.

If we choose build new image and test it, it need much time. It should be a long term for fix the issue.

So based on above difference, I prefer pin to 0.2.0

[Steven Dake (sdake)]

One possible reason is if the modules were built via zfs, fedora doesn't support zfs. So perhaps that is the cause.

Rebuilding the images changes everything and is a big effort. My devstack is broken atm, and I need to get it fixed before I'd be willing to take on a rebuild of the images.

[Andrew Melton (andrew-melton)]

I've confirmed the issue with the swarm image on a plain Fedora 22 with it's Docker 1.6.0. If I run 'docker run --rm -v /tmp swarm', it'll work, just like the openshift folks do here: https://bugzilla.redhat.com/show_bug.cgi?id=1223662. It's definitely an issue with Fedora's custom 1.6.0 release. Same thing with the docker.io/cirros issue, I only get that when running with Fedora's custom 1.6.0. Docker's F22 1.7.0 release seems to work just fine.

[Andrew Melton (andrew-melton)]

I've moved the fedora 22 issues to this bug to track them separately: https://bugs.launchpad.net/magnum/+bug/1468881

[OpenStack Infra (hudson-openstack)]

Change abandoned by Andrew Melton (<email address hidden>) on branch: master
Review: https://review.openstack.org/195716
Reason: Abandoning in favor of https://review.openstack.org/#/c/194963/